Match Explanation

# Recipe 05 — Candidate-job match explanation

Given a structured candidate record and a structured job record, explain in five bullets *why* this match works (or does not). This is the explainer layer — the score itself should be computed in code, not in the model.

## When to use

You have a deterministic match score (computed by your platform or a simple weighted formula) and you want the model to translate it into human language for a recruiter or hiring manager.

## Why explanation, not scoring?

Models are not reliable rankers. They are excellent narrators. The pattern that works:

1. **Compute the score in code** — weighted skill overlap, seniority match, location compatibility, salary delta. Deterministic. Cacheable. Auditable.
2. **Ask the model to explain that specific score** — given both records and the score, generate the human-readable rationale.

This is the same approach Fitlane AI uses in production. See [How match scoring works in Fitlane AI](https://fitlaneai.com) (link will become live with the product launch).

## System prompt